Egocentric video summarization based on video frame quality assessment

Abstract

In this paper, we present an egocentric video summarization system based on video frame quality evaluation. Our evaluation criteria is motivated by the fact that high quality frame deliver more useful information than low quality video and therefore better in representing video summary index. We measure video frame quality by measuring the amount of motion blur, edge compactness, colorfulness, and the existence of human face. To avoid selecting similar frame repetitively, we use relative score evaluation to select locally the best quality frame as an index frame. We can also combine our system with another video summarization system to select a better index frame given the index frame from the other system. Our experiments show that our system is effective and our user study confirms that our selected video frames are more preferable to users.